Thoughts on Mark Lewis’ Future Storage

EMC’s Mark Lewis posted another thoughtful “blog episode”, outlining five predictions he has for the next few years. I don’t really agree with him much more than I did the last time, but it’s an interesting read nonetheless!

1. Offline Storage becomes extinct for most uses

Mark claims that “he cost/availability of network bandwidth, the cost of the people side of tape storage and handling, disk cost declines, multi-site DR, data de-duplication and many more factors” will kill tape eventually. Let’s take some of these in turn…

Network bandwidth, DR, and related tech

Like Hu Yoshida, who claimed we home users would rely on service providers rather than have a terabyte in our home, Mark assumes plentiful and cheap bandwidth. But this is simply not the case for many organizations – bandwidth is one of the biggest IT costs, especially for small companies, and other issues like latency and availability abound. Even with technical fixes like WAFS compression and acceleration, many organizations will not be able to replicate much of their storage for the time being. Unless there is some new blossoming of bandwidth, I don’t see this changing.

Disk cost declines

Yes, disk costs are going down. But tape costs are, too! Today’s tapes are still 1/3 the cost of the same space on disk and have remained thus for the last decade. Tape drives are still expensive, but considering that most organizations still use a “weekly full” schedule, the cost of media becomes their big concern, and tape has disk beat hands-down. Unless we all give up on daily backups for data recovery and start using snapshots, tape will not die. And, although I fervently believe in the rightness of this snapshot-based strategy, I doubt it’ll happen for some time to come.

People costs

This is always the problem with backup systems. Not so much the cost of people, but the reliability of human tape handling. Pour in enough money to make tape handling reliable and you’re pouring a lot into it indeed! But people costs are notoriously hard for IT to recognize, let alone manage or reduce. I don’t see this killing tape any more than it could kill the multitude of other time sinks in the IT world.

2. Flash becomes a viable Tier 1 storage option

Didn’t I just cover this the other day? Here’s the summary – solid-state disks are still way too expensive and way too short-lived for enterprise use. Put some cache in front of a disk and forget SSD. Note to Mark: This is exactly what EMC did in 1992 when they turned the Orion into the Symmetrix.

As for the OLTP angle – I agree that this one application of storage technology needs high performance. And I agree that NAND flash is cheaper than RAM and other technologies. Maybe use NAND as a cache? Or hybrid drives? But this is only a small component of enterprise data.

3. High Capacity/Low Cost Disk becomes the principle “bulk storage” medium

I can’t argue this one. Big fat slow disks are here in the enterprise space and are here to stay. Auto-magic virtualization tech will put the right data on them, and we’ll all hold hands and sing.

4. FCoE SANs become the FC evolution path for OLTP storage

Let’s see – Mark is predicting that FCoE beats FC and iSCSI in the high-end enterprise market. Well, I suppose you could claim that future versions of FC will run over Ethernet hardware, but FCoE?!? That dog is just now getting out of bed – let’s give it more than 3-5 years to rule the world, ok? Oh, and iSCSI still works great to let’s not count it out!

5. Web Storage Applications move away from SCSI and File System protocols and become connected principally via “Object” protocols (e.g. SOAP, REST)

I really wish this would happen. Mark cites Amazon.com’s S3 and EMC’s Documentum as examples of object stores. I think this prediction is really up in the air. I’m glad he didn’t specify exactly which object storage protocol will win, because I suspect it’ll be something we haven’t heard of yet. But I think he’s right that increasingly modular and relational applications found in this “Web 2.0” world will definitely prefer object storage instead of plain filesystems. Hopefully, enterprise applications will follow, and we’ll all laugh about the old days when we used to try to manage data with no metadata…

So I guess I don’t agree with Mark on a lot of these predictions. I really like what he has to say, for the most part. But it just doesn’t seem realistic to expect them all to come about in 3 to 5 years.

But hey, as they say, opinions are like belly buttons – everyone has one. Even me! Well, actually, I hear “they” use a different body part in this idiom, but this is a family blog!